NFS stale mounts when exporting system substituted by another box.

[Derek Whigham_1]

Scenario: need to replace server with a clone box. Replacment box exports some filesystems upon startup, AND remote clients succeed in mounting them, BUT attempts to read (world-readable) files in the cross mounted directories fail.

Cycle of exporting box - no difference
Cycle of automounting box - no difference

Error is 0x46 (70) i.e. ESTALE

[Dietmar Konermann]

Basically an HP-UX NFS server builds up the file handle for a give export using the triple major#/minor#/inode#.

The client gets that handle at during mount time. So umount/mount on the client side should resolve the issue. It's surprising that power-cycling the client did not help.

Best regards...
Dietmar.

[Steven E. Protter]

I would take the following steps on the new server.

verify all entries in /etc/fstab are correct

verify permisions on the exported filesystems

verify the /etc/exports file

exportfs -av

That will give you verbose output and possibily more things to do on export.

Then retry the mount.

SEP

[stuart mcguinness]

thanks for postings; got my own reg now

should also have said that "clone" box has only / and /stand filesystems (72Gb) but that original had multiple LVM partitions,

/
/home
/usr
/var
/opt

and all these are exported to 9 or 10 named client hosts which barf when they can't read a necessary file over NFS

Now the /etc/exports file was same on both old box and clone and I am certain that the client cycle did not recover situation.

Tried changing exports to have / only as a single fs corresponding to "clone" actual LVM status. Not certain whether a server & client bounce was attempted at this point. This will be next thing to try.

Note that permissions and traverse permissions are correct on target directories & files, that root user on client side CAN see the files and directories and only fails when trying to "cat", "vi" or "more" the (perms 644) desired files. When it's all working, root maps to uid -2 over mount but CAN still read xx4 files.

Do you think that having multiple mounted directories (not separate filesystems) would lead to client nfsd retaining inode info from previous mounted filesystems and therefore not recognising that /opt /var etc are under the new, "/" only, export?

Seems likely, but I want to have a plausible scenario before trashing customer system again!!

[Derek Whigham_1]

Thanks for the Help